A new study combines interpretable machine learning with laboratory experiments to uncover how the elemental composition of biochar influences persistent free radicals, offering a potential tool for predicting biochar reactivity and environmental risks.
Biochar, a carbon-rich material produced by heating biomass under oxygen-limited conditions, is widely studied for applications ranging from soil improvement to pollution control. Yet biochar can also contain persistent free radicals, or PFRs, which may remain stable for days or months. These radicals can help generate reactive oxygen species that break down pollutants, but they may also contribute to oxidative stress in living organisms.
Understanding what controls these radicals is therefore important for designing biochar that is both effective and environmentally responsible.
Researchers from Kunming University of Science & Technology analyzed published data using six machine-learning approaches, including XGBoost, gradient boosting, support vector regression, a shallow neural network, random forest, and ensemble learning. They then used interpretable machine-learning tools to determine which elemental properties most strongly influenced PFR concentration and radical type.
The analysis identified the hydrogen-to-carbon ratio, or H/C, and oxygen content as the most important descriptors of PFR concentration. Oxygen content and the oxygen-to-carbon ratio, or O/C, were the strongest predictors of the g-Factor, a measurement used to distinguish different types of free radicals.
"Our results show that relatively simple elemental information can provide valuable clues about the behavior of persistent free radicals in biochar," said Wenmei Tao, corresponding author of the study. "Combining interpretable machine learning with experimental validation may help us better understand how these radicals form and ultimately support more informed biochar design and environmental risk assessment."
The researchers assembled 263 paired records of PFR concentration and g-Factor from previously published studies. The best-performing models achieved test R² values of 0.7797 for PFR concentration and 0.7647 for g-Factor. The models were also highly effective at distinguishing samples with relatively high and low values, with ROC-AUC values exceeding 0.96.
Interpretable analyses provided further insight into the relationships behind those predictions. Lower H/C values and lower oxygen content were generally associated with higher PFR concentrations. In contrast, increasing oxygen content and O/C tended to increase the g-Factor, indicating a shift toward oxygen-centered radicals.
These patterns suggest that changes in elemental composition during biomass pyrolysis are closely connected to both the amount and type of persistent free radicals formed in biochar.
To test whether the machine-learning findings held up experimentally, the researchers produced biochars from cellulose, lignin, peanut hull, rice straw, and pine sawdust at different pyrolysis temperatures. Independent measurements supported the major trends predicted by the models. Spearman correlation analysis confirmed that H/C and oxygen content were most strongly associated with PFR concentration, while oxygen content and O/C showed the strongest relationships with g-Factor.
The researchers also found that when O/C values were similar, higher hydrogen content was associated with a higher g-Factor, revealing an additional interaction that would be difficult to identify using simple linear relationships alone.
The study provides a data-driven framework for linking readily measurable elemental properties to the behavior of persistent free radicals in lignocellulose-derived biochar. The findings could help researchers better evaluate PFR-related environmental risks and guide the development of biochars with properties tailored for environmental remediation.
===
Journal reference: Gao L, Xu C, Li M, Xu Z, Tao W. 2026. Elemental composition-based prediction of persistent free radicals concentration and g-Factor in lignocellulose-derived biochar combining interpretable machine learning and experimental analysis. Biochar X 2: e022 doi: 10.48130/bchax-0026-0020
https://www.maxapress.com/article/doi/10.48130/bchax-0026-0020
===
About the Journal:
Biochar X (e-ISSN: 3070-1686) is an open access, online-only journal aims to transcend traditional disciplinary boundaries by providing a multidisciplinary platform for the exchange of cutting-edge research in both fundamental and applied aspects of biochar. The journal is dedicated to supporting the global biochar research community by offering an innovative, efficient, and professional outlet for sharing new findings and perspectives. Its core focus lies in the discovery of novel insights and the development of emerging applications in the rapidly growing field of biochar science.
